coaction can now save the logs in chunks to reduce the memory usage. This is an alternative to np.memmap (which is not currently supported) and not a replacement. np.memmap support is planned be in a future release.
0.0.6
Implemented agents in coaction now has the option the select the logged parameters.
0.0.5
coaction now has an individual Q-learning implementation. This release also has minor improvements and bug fixes.
0.0.4
This is the first release of the project.
coaction can run parallel simulations for learning in Markov games. It currently includes fictitious play variants and custom agent support.